This action might not be possible to undo. Are you sure you want to continue?

# Sense and Proof

Carlo Penco, Daniele Porello

1

Frege, Carnap and the contrast between cognitive semantic sense

In a paper given at the 1997 SILFS Conference, [Pen98] claims that a strong tension between a semantic and a cognitive notion of sense was already present in Frege’s writings. Many authors have discussed recently this tension in Frege, but few of them remark that Carnap was probably the ﬁrst to attempt to compose the tension. Carnap ([Car47]) soon realized that his conception of intension, although suitable to treat a semantic notion of sense as truth condition on the lines developed by Wittgenstein, was not enough to solve the problem posed by Frege on belief contexts. He devised therefore the notion of intensional structure and intensional isomorphism, as a more ﬁne-grained notion that the semantic notion o intension and intensional equivalence. For many reasons however his notion did not appear successful. In this short paper we try to develop a new perspective which tries to compose the tension between a semantic and a cognitive notion of sense, suggesting the use of proof theory as a mean to interpret the cognitive notion of sense. In the Fregean view, sense is something which is there to be grasped from a speaker. The concept of grasping a sense or understanding is co-original with the notion of sense. However it is not clear in Frege what is the object of understanding. Frege himself made interesting remarks on the limitations of understanding complex mathematical formula with all their connections with other parts of mathematics. In his famous argument on the “intuitive diﬀerence of thought” (as it has been named by Gareth Evans) Frege claims that if it is possible to understand two sentences, and coherently believe one and disbelieve the other, then those sentences express diﬀerent senses or thoughts. The Fregean example is the belief that Hesperus is a planet and Phosphorus is not a planet, given by the ignorance of the identity Hesperus=Phosporus. A lack of knowledge permits a person who hold erroneous beliefs to be considered rational, because the erroneous beliefs express two

given that most of his examples -given in his last writings on logic . semantic notion clashes with the ﬁrst if we try to apply also to the semantic notion the criterion of the intuitive diﬀerence of thoughts. while understanding the basic notions used. with a limited understanding. if we may believe two sentences as having diﬀerent truth value because of lack of information. 2 Understanding sense as truth conditions Let us begin with an analysis of what it means “to understand sense as truth conditions”. to what she grasps. to her limited accessibility to the information. composing the original tension. The criterion links sense identity to what a subject believes. but there are many passages where Frege accepts the idea of limited computational capacity as a way to explain why two expressions expresses diﬀerent senses. Therefore she may believe that A → B is true while ¬(A ∧ ¬B) is false. We claim that the ambiguity in Frege’s concept of sense depends on the lack of logical instruments that has been developed later with proof theory.deal with elementary logical equivalences between connectives. The semantic notion of sense is linked to the criterion of logical equivalence: in an example proposed by Frege (letter to Husserl 1906. therefore the two sentences would express .2 Carlo Penco. And we think that proof theory may help to deﬁne diﬀerent levels of understanding. The idea of sameness of sense given by logical equivalence is coherent with the concept of sense as truth conditions made by Wittgenstein in his Tractatus. the understanding .contrary to what Frege says two diﬀerent (cognitive) senses. in a way which is diﬀerent from the original suggestions given by Dummett with the choice of a veriﬁcationistic theory of meaning. the meaning of the connectives. This possibility is enforced by what Frege says in a later paper on negation. see [Fre76]. The second. Daniele Porello diﬀerent thoughts. why shouldn’t we accept the same uncertain attitude towards two sentences with the same truth conditions. when he assert that A and not not A express diﬀerent thoughts. We may imagine a person who is so slow that she cannot realize that two logical equivalent sentences produce the same truth tables. Frege probably connects this idea with the criterion of immediate recognizability. the working of the symbolism. Here the limitation is given by empirical information. When we consider complex sentences. whose truth is not evident to the believer. but such that we accept one of them while remaining uncertain on the truth condition of the other? Actually we may lack computational capacity.) two sentences which express a logical equivalence like A → B and ¬(A∧¬B) express the same thought. in this way we may distinguish the semantic and the cognitive aspect of sense.

the subjects immediately believes all the logically equivalent sentences. First. the sentence is true. then we are forced to say that the subject believes that (2) is also true. Suppose for example that a subject. then she believes that (A ∧ ¬B) is true. she believes that it is the case that A is false or B is true. and since these are precisely the truth-conditions of ¬(A ∧ ¬B) and the subject knows that if these condition are satisﬁed then ¬(A ∧ ¬B) is true (since she understands both sentences). no matter how complex they are. by deﬁnition of material implication. we cannot concede the possibility that a subject might believe (1) true and (2) false. then he must believe all the sentences that are logically equivalent to A. if the meaning of the conjunction and the negation is known. But if she believes that ¬(A ∧ ¬B) is false. she knows that A → B is true when either A is false or B is true. when we know for which values of the atomic propositions occurring in A. With our notion of understanding sense as truth conditions we are compelled to make our speaker logically omniscient. We have no means to say that she doesn’t realize that she is contradicting. then she believes that A is false or B is true. If a subject understands A → B then. it follows that if a subject believes a sentence A. A is true and B is false). If we do not consider how subjects grasp those truth conditions. So the truth conditions of the sentences involved are the same. we have two logically equivalent sentences (1) A → B (2) ¬(A ∧ ¬B). then she knows that the sentence is true when it is false that A and B is true. then we face the following problems. since we cannot deny it without contradicting our deﬁnitions. since she believes (1) true. believes that A → B is true and believes that ¬(A ∧ ¬B) is false. If the subject understands ¬(A ∧ ¬B). since we just proved it as a fact simply entailed by our hypotheses. . Moreover.Sense and Proof 3 of truth conditions amounts to knowing the corresponding truth tables. we cannot represent such situation. Then. If a subject believes that A → B is true. Even in this case. Consider the example considered above. that he knows that ¬(A ∧ ¬B) is true when either A is false or B is true. so she believes that A is true and B is false. the only way in which we may claim that she believes (1) true and (2) false is claiming that she is accepting a contradiction. We may however ask whether it is possible that a subject believes that (1) is true while having no opinion about the truth value of (2). for example. From this argument. if we keep the hypotheses at issue. which entails. Since the information the subject should manage is not coherent (A is false or B is true. by mistake or logical confusion. But why does she believe it? Simply because of our deﬁnitions. The sense of a complex sentence A is then known.

even if we try to employ a cognitive notion of sense (e. A proper notion of cognitive sense should be grounded not on a strong notion of understanding requiring full grasping of truth conditions. which seem to be the relevant aspect in a cognitive notion of sense. The notion of sense of a sentences in mere terms of truth conditions fails to capture all the information concerning the complexity of the process of grasping the truth condition. or computational. and moreover accepts (1) A → B. namely the notion of sense as computing procedure hinted at by Frege (see [fre93]. what does she need in order to accept also (2) ¬(A ∧ ¬B)? The point is that if we aim to describe a cognitive notion of sense.4 Carlo Penco. point of view. we can state the problem at issue with the following question: (Q) Assuming that a subject understands (1) A → B and (2) ¬(A ∧ ¬B). 4 Limited knowledge and procedures Assuming that understanding a formula is not necessarily understanding directly its truth conditions. we note that the formula ¬(A ∧ ¬B) requires more calculation than A → B. It seems that if we assume the deﬁnition of understanding a sentence as mere grasping truth conditions. but understanding its mode of composition and the meaning of the connectives. g. we cannot consider the process of understanding of the two sentence as imme1 See [Pen03b] . We will therefore need to enrich the notion of sense considering another aspect of sense that seems more suitable to deal with cognitive aspects.). In the following sections. we will advance a proposal for deﬁning a cognitive notion of sense which is coherent and compatible with an objective one. Daniele Porello 3 A weaker notion of understanding The logical omniscience of a subject described by the assumptions we made concerning the notion of understanding a sentence is completely useless from a cognitive. the one presented by the immediate recognizability criterion) we are not allowed to deﬁne a cognitive diﬀerence between logically equivalent sentences. If however we look closer at the same argument we mentioned to ﬁnd out the truth conditions of the sentences involved. but on a weaker notion of understanding based on the idea of limited knowledge (or even on bounded rationality) 1 .

However. We need to consider the process of understanding. provided they have a proof theory and a truth values semantics. We are not claiming that it is always the case that someone who accepts a sentence is able to justify it showing a proof. The problem (Q) therefore may be generally solved in our setting saying that a subject can understand (1) and (2) since she can manage a procedure to grasp their truth conditions. 5 Proof theory for representing procedures We sketch a formal setting for representing the deﬁnitions we proposed which allows us to keep both a classical truth conditional style semantics and [Dum91]. and then it is committed with intuitionistc logic. or a computation. our proposal is to keep the truth conditional approach – since it allows to state clearly the relationship between a cognitive notion of sense and an objective.4 It is useful to remark that the approach we are suggesting is not to be intended as a representation of the explicit knowledge speakers have. 3 See 2 see . The notion of proof. is the notion of proof which may be deﬁned within some suitable logical calculus. say (2) given (1). since it is not decided yet which formal calculus is adequate to represent semantic understanding. A good way to represent procedures. since the subject may fail to manage the procedure of detecting the logical relationship between (1) and (2). p. [Dum79]. Here we are presenting an approach which could be applied in diﬀerent logical calculi. has been applied for example by Michael Dummett 3 to deﬁne his justiﬁcation semantics.in case she believes one true and the other false . or semantic. We are rather suggesting that the notion of proof is a useful tool for representing implicit knowledge5 speakers show when they understand sentences. as we shall see in more detail in the next section. rather than proposing an alternative semantic theory based on diﬀerent key concepts. but . as a process mediated by a procedure. Therefore we reformulate the truth conditional approach considering the procedure of grasping truth conditions as a constitutive feature of understanding sentences. 4 The justiﬁcation semantics proposed by Dummett uses the notion of proof as semantic value.Sense and Proof 5 diate 2 . Our basic deﬁnition is then that a subject understands a sentence when she can perform a procedure of grasping truth conditions of the sentence.she may fail to realize she reaches a contradiction. 5 We refer to [Dum91]. Otherwise we would lose the information concerning the complexity of the process of understanding sentences which is essential for a cognitive notion of sense. notion of sense – enriching it by means of a notion of procedure. or more generally the notion of justiﬁcation. 139.

namely the cognitive competence of a subject amounts to manage a subclass of the rules of inference which are deﬁned in a logical calculus. (C) The cognitive sense of a sentence A is a class of rules a subject manages. we don’t want to assume that he is able to draw the conclusion at any degree of complexity: we are not assuming she can 6 An interesting choice would be to state our deﬁnitions within linear logic (see [Gir06]). linear logic allows to deﬁne where resources are actually needed to be bound and where we can assume an unbounded number of tokens. We propose to use proof theory to represent the procedure of grasping truth conditions. Moreover. the partial comprehension can also be stated in terms of complexity bounds on the application of those rules. and it may also be considered as an analysis of the properties of classical and intuitionistic proofs. Remark that in this way we may keep a notion of co-tenability of thoughts. Brieﬂy. This aspect is crucial in order to go into the relationship between a cognitive and a semantic notion of sense. 7 The fact that we don’t allow individual or private strategies aims to keep some features of the Fregean anti-psychologism: the sense of a sentence doesn’t depend on the representation nor depends on private aspects of the comprehension of meaning. We leave a deeper examination of this approach for further work.6 Carlo Penco. in this way we can describe the failure of detecting logically equivalent sentence as a lack of logical competence due to the complexity of the sentences involved. Here we will not refer precisely to a particular logical calculus. . it is not clear if we can consider the semantics of linear logic. Daniele Porello to account for the complexity of the procedure of grasping truth conditions. which lead to grasp (partial) truth conditions of A. which states some important intuition about the relationship between the meaning of a sentence and a subject who understands it. linear logic has been applied to deﬁne formal grammars for natural languages working both for syntactical aspect of sentence understanding and for composition of meanings (see for example [Mor94] and [Car97]). For example. However. which lead to the truth conditions of A. It is important to remark that the partial understanding we are deﬁning depends on subjects just in the sense that subjects manage some of the rules required to build a proof of a given sentence. in the sense that both can be embedded in linear logic. Linear logic may be considered more general than intuitionistic or classical logic. based on the algebraic structure of phase space (see [Gir87]). So the relationship between (C) and (S) can be stated in terms of an inclusion. (S) The semantic sense of a sentence A is the whole class of rules deﬁning a proof of A. Moreover. It doesn’t mean that a subject has individual or private rules for building proofs7 . if we assume that a subject is able to perform a modus ponens. rather we will propose some general idea which may be applied in diﬀerent logical frameworks6 We consider two notions of sense: a semantic one and a cognitive one. as a truth value semantics: we would need in particular to investigate which notion of truth is formalized by that structure.

Suppose a subject understands (1) and (2).Sense and Proof 7 perform a proof consisting in an unbounded iterated or nested applications of modus ponens. represented by the logical inference: (π): B A→B So the subject knows that she is accepting A → B. Therefore. in order to get truth conditions for (2). Consider again Frege example. even if we try to deﬁne a partial understanding by means of the immediate recognizability criterion.8 We consider some example using natural deduction for classical logic. in the sense that we do not need to assume subjects are able to grasp all the possible case described by a truth table (in this example. the class deﬁning the cognitive notion will be a sub-class of the class involved in the objective notion of sense. she considers (1) true. as we saw. for example by means of a proof like: (π ): [A ∧ ¬B]1 ∧E A B π A→B →E ⊥ ¬(A ∧ ¬B) [A ∧ ¬B]1 ∧E ¬B ¬E ¬ I. a subject may be able to manage the following procedure. we are led either to make the subject contradict himself or to make the subject logically omniscient. In this way we may represent a partial comprehension of truth conditions. a subject may be able to perform the following procedure: (π ): ¬A ¬(A ∧ ¬B) Which doesn’t entail a complete knowledge of truth conditions of (2). since she is accepting B. . 1 8 Remark that it is diﬃcult to speak of partial understanding using mere truthconditional deﬁnition of sense since. Moreover. In order to get the truth conditions for (1) A → B. So we can assume that a subject can understand (1) and (2) and she can accept (1) but she can fail to accept (2) simply because she can fail to manage the procedure of detecting the connection between (1) and (2). by means of the procedures we mentioned. that A → B may be true also if A is false). We can represent what the subject needs in order to accept also (2). Moreover.

We chose this example. therefore we may say that a subject who accepts one of the two sentences is not rational if he doesn’t accept the other. so we can apply the rule of introducing negation (¬I) and discharge hypotheses marked by 1. We consider commutative use of conjunction. Consider the problem of quantiﬁer-scope ambiguity. . Actually we cannot deﬁne a normal form procedure without considering empirical data concerning subjects eﬀective performance. Eliminating conjunction. Daniele Porello Here π represents the procedure the subject can perform in order to understand and then accept (1). Consider now an example showing a kind of same level complexity and ask again the same question concerning what a subject needs in order to detect logically equivalent sentences. In this case. A subject can manage the procedure π that leads her to accept (1). we obtain A and we obtain ¬B. then she has to accept also B ∧ A. besides a common logical content. depending on the narrow (∀∃) or wide (∃∀) scope. If she accepts A ∧ B. in order to accept (2) a proof is required. May a subject accept A ∧ B while not accepting B ∧ A? If she understands both sentences. But if this metalinguistic use of “and” is commutative. Assume by contradiction. We used a natural deduction proof just to show an example. may be a staring point in order to deﬁne classes of meanings sharing a same measure of such complexity. then she grasps in a certain way a procedure represented by the rules for the conjunction. since she can manage both procedures. then she should accept both A and B. we obtain by eliminating conditional B. 10 The aim of this proof theoretic approach to the complexity of understanding meaning would be a sort of normal form for the proof representing procedures of grasping truth conditions.9 So we can use the proof π to represent the process of accepting (2) given (1). The sentence “Someone loves everyone” allows two diﬀerent readings. so to give a prooftheoretical account of the criterion of immediate recognizability. we have two formulas which share a same level of understanding complexity. 10 We conclude mentioning an example taken from the literature on the application of proof theoretical notions in formal semantics which show how proof theory can represent subjects’ diﬀerent performances also in case of quantiﬁed sentences11 . A careful examination of 9 We can read the proof in the following way. just not to be committed a priori with classical logic. From hypothesis (1) and from A. Actually the distinction between partial and full understanding stated in terms of proof can be reformulated for semantic theory that insist on other notion of semantic value. 11 See [Mor00]. we could have chosen other proof theoretical calculi.8 Carlo Penco. if we are in a commutative framework. Of course our approach should take into account data describing subjects eﬀective performances. In this way it is possible to argue about complexity bounds to put on the process itself. The notion of normalization of proof. which is a central issue in proof theory. and we used the intuitinonistic rule for negation. But B and ¬B entail a contradiction (⊥). A ∧ ¬B.

in case of two logically connected sentences. 1976. a subject can access a procedure for grasping the truth conditions of the sentence. Elements of Intuitionism. Meaning and necessity. BIBLIOGRAPHY [Car47] Rudolf Carnap. which is the peculiar proof theory for linear logic. 6 Conclusion The proposal we presented allows to consider both a cognitive and a semantic notion of sense and. What is a theory of meaning (ii). [Car97] B.). Truth and Meaning. Summing up. calculation or other means. as it happens for example in many attempts to conciliate those two aspects of Fregean notion of sense. or proofs. while the cognitive notion of sense is deﬁned as a partial access to semantic sense. we can claim that it is rational. MIT Press. This approach points at a more sophisticated notion of rationality which can include the process of learning new procedures (for example. Carpenter. we can see how the two notions interact: they are not distinct features. The semantic notion of sense is given by the whole class of procedures. of Chicago Press.Sense and Proof 9 proofs representing the meanings of those sentences12 it is possible to express the fact that the preferred reading (∃∀) has a lower complexity degree than the other. the representation at issue employs proof nets. 12 Actually. or better it is possible without contradiction. So we provided the theoretical possibility for suspending judgement until. In Evans and McDowell. We stated the relationship between cognitive sense and semantic or objective sense in terms of a partial understanding speakers have of meaning. to accept a sentence while having no opinion concerning the other. this is the most interesting point. editors. or should have. 1997. 1947. [Dum77] Michael Dummett. Moreover we used a general notion of procedure. U. So it seems that the opposition between a truth-conditional semantics and a justiﬁcation-semantics (Dummett) may be weakened considering the proof as a way to grasp truth conditions of a given sentence. represented by the notion of proof in some suitable logical system. of the given sentence that give the truth conditions of the sentence. Oxford University Press. when the complexity of the logically connected sentences is diﬀerent. Type Logical semantics. 1977. Cambridge (Mass. Oxford University Press. by means of interaction) rather than considering rationality as static set of features subjects have. In this way it is possible to develop a quite precise notion of cognitive relationship between subject performances and meaning of a sentence. by reﬂection. . [Dum76] Michael Dummett.

Dordrecht. Tr. [Gir87] Jean-Yves Girard. Moschovakis. 24(2):87–109. Lecture Notes in Logic. Prospettive della Logica e della Filosoﬁa della scienza: Atti del Convegno SILFS. Yves Lafont. Rist. 9:53–65. Linear logic. Uber Sinn und Bedeutung. Review of Modern Logic. In Proceedings of the 1993 ASL Summer Meeting in Helsinki. The Logical Basis of Metaphysics. 1993. Meaning in terms of justiﬁcation. Margalit. Topoi. Morrill. [Gir06] Jean-Yves Girard. 1989. Frege e carnap: verso una teoria integrata del senso. ed. 1998. vol I. [Pen03b] Carlo Penco. Proof theory and meaning. 100:25–50. MIT Press. ¨ [Fre92] Gottlob Frege. [fre93] Gottlob frege. two senses. Type logical Grammars. Meaning and Use. Incremental processing and acceptability. Reidel. Cambridge. 1993. Cambridge Tracts in Theoretical Computer Science 7. Reidel. [Pen03a] Carlo Penco. [Hey56] Arend Heyting. Frege. Hermann. [Mor94] G. sense and limited rationality. In Roberto Cordeschi Vincenzo Fano Vito Michele Abrusci. Hildesheim. 1995. 2006. Engl. 50:1–102. M. [vB95] J. editors. 1986. Engl. Dordrecht. Felix Meiner. The Basic law of Arithmetics Univ. 1956. Jena. [Mor00] Glyn Morrill. Harvard University Press. [Pen98] Carlo Penco. Kaal.. Intuitionism. Gabbay and F. Computational Linguistics. [GLT89] Jean-Yves Girard. 2002. [Dum91] Michael Dummett. In D. Taylor. editor. 1980. Lambdas and Dynamic Logic. pages 471–506. of California Press. Sense and denotation as algorithm and value. 1987. ETS. . Categories. 21:11–19. Proofs and Types. 1976. Dordrecht. Pisa. Reidelm. and P. 2000. Hamburg. Cours de Logique I. by B. Wissenshaftliches Briefwechsel. [Sun86] Goran Sundholm. Amsterdam. What does the appeal to use do for the theory of meaning? In A. 1892. Olms. van Benthem. [Dum02] Michael Dummett. Handbook of Philosophical Logic: Volume III: Alternatives to Classical Logic. Pohle. 1991. 1979. Le Point Aveugle. pages 345–360. [Fre76] G.10 Carlo Penco. Furth (ed. Frege: Two theses. North-Holland. Philosophical and Mathematical Correspondence. Daniele Porello [Dum79] Michael Dummett. Hamburg. Language in Action. Springer-Verlag. Zeitschriﬀt f¨r Philosophie und u Philosophische Kritik. Grundgesetze der Arithmetik. Carlo Cellucci. Cambridge Mass.) Frege. 2003. 1994. Vers la Perfection. tr. Oxford. Theoretical Computer Science. Guenthner. [Mos93] Y. History anf Philosophy of Logics. 1893. 1964. Berckeley. 2003. editor. Blackwell. 26(3):319–338. Cambridge University Press. Mc Guinness and translated by H. Frege. Paris.. 1962.